Letter doi:10.

1038/nature24291

Optical emission from a kilonova following a

gravitational-wave-detected neutron-star merger
Iair Arcavi1,2, Griffin Hosseinzadeh1,2, D. Andrew Howell1,2, Curtis McCully1,2, Dovi Poznanski3, Daniel Kasen4,5,
Jennifer Barnes6, Michael Zaltzman3, Sergiy Vasylyev1,2, Dan Maoz3 & Stefano Valenti7

The merger of two neutron stars has been predicted to produce
an optical–infrared transient (lasting a few days) known as a
‘kilonova’, powered by the radioactive decay of neutron-rich species
synthesized in the merger1–5. Evidence that short γ-ray bursts also
arise from neutron-star mergers has been accumulating6–8. In
models2,9 of such mergers, a small amount of mass (10−4–10−2
solar masses) with a low electron fraction is ejected at high velocities
(0.1–0.3 times light speed) or carried out by winds from an
accretion disk formed around the newly merged object10,11. This
mass is expected to undergo rapid neutron capture (r-process)
nucleosynthesis, leading to the formation of radioactive elements
that release energy as they decay, powering an electromagnetic
transient1–3,9–14. A large uncertainty in the composition of the newly
synthesized material leads to various expected colours, durations
and luminosities for such transients11–14. Observational evidence
for kilonovae has so far been inconclusive because it was based on
cases15–19 of moderate excess emission detected in the afterglows of
γ-ray bursts. Here we report optical to near-infrared observations
of a transient coincident with the detection of the gravitational-wave
signature of a binary neutron-star merger and with a low-luminosity
short-duration γ-ray burst20. Our observations, taken roughly every
eight hours over a few days following the gravitational-wave trigger,
reveal an initial blue excess, with fast optical fading and reddening.
Using numerical models21, we conclude that our data are broadly
consistent with a light curve powered by a few hundredths of a solar
mass of low-opacity material corresponding to lanthanide-poor (a
fraction of 10−4.5 by mass) ejecta.
GW170817 was detected22 by the LIGO23 and Virgo24 gravitational-
wave detectors on 17 August 2017 at 12:41:04 (universal time
(ut) is used throughout; we adopt this as the time of the merger).
Approximately two seconds later, a low-luminosity short-duration
γ-ray burst, GRB 170817A, was detected25 by the Gamma-ray Burst
Monitor (GBM) on board the Fermi satellite. A few hours later, the
gravitational-wave signal was robustly identified as the signature of
a binary neutron-star merger 40 ± 8 Mpc away in a region of the sky
coincident with the Fermi localization of the γ-ray burst26 (Fig. 1).
Shortly after receiving the gravitational-wave localization, we
activated our pre-approved program to search for an optical coun-
terpart with the Las Cumbres Observatory (LCO) global network of
robotic telescopes27. Given the size of the LIGO–Virgo localization
region (about 30 square degrees) compared to the field of view of
our cameras (about 0.2 square degrees), our search strategy involved
targeting specific galaxies28 (chosen from the GLADE catalogue;
http://aquarius.elte.hu/glade/) at the reported distance range

45°
a 30° b
15°

0°

–15°

–30°
180°

–45°
240°
120°

–60°

–75°
300°
60°

Figure 1 | Localizations of the gravitational wave, the γ-ray burst and (for example, 240°) and declination (for example, −30°) values are shown.
the kilonova on the sky. a, Our localization of the kilonova AT 2017gfo is The position of the Sun is indicated by the symbol . b, A more detailed
shown by the blue filled circle, together with the localization of GW170817 view of the kilonova region. Empty circles indicate the locations of other
(blue contours)26 and that of GRB 170817A (red contours)25. The contours galaxies searched by our LCO follow-up program36.
indicate 1σ, 2σ and 3σ confidence bounds. Representative right ascension
1
Department of Physics, University of California, Santa Barbara, California 93106-9530, USA. 2Las Cumbres Observatory, 6740 Cortona Drive, Suite 102, Goleta, California 93117-5575,
USA. 3School of Physics and Astronomy, Tel Aviv University, Tel Aviv 69978, Israel. 4Nuclear Science Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720-8169, USA.
5
Departments of Physics and Astronomy, University of California, Berkeley, California 94720-7300, USA. 6Columbia Astrophysics Laboratory, Columbia University, New York, New York 10027, USA.
7
Department of Physics, University of California, 1 Shields Avenue, Davis, California 95616-5270, USA.

0 0 M o n t h 2 0 1 7 | VO L 0 0 0 | NAT U R E | 1
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH Letter
DSS LCO
18 August 2017 00:15:23
N
10″
E
Figure 2 | LCO discovery image of the kilonova AT 2017gfo in the
galaxy NGC 4993. The w-band LCO image (right), centred on NGC 4993,
clearly shows a new source (marked with white ticks) compared to an
archival image (left) taken on 9 April 1992 with the RG610 filter as part
of the Anglo-Australian Observatory Second Epoch Survey (AAO-SES),
retrieved via the Digitized Sky Survey (DSS).
and location area included in the LIGO–Virgo three-dimensional
localization29 (see Methods).
The fifth galaxy on our prioritized list was NGC 4993, an S0
galaxy 39.5 Mpc away30. We observed it with one of the LCO 1-m
telescopes at the Cerro Tololo Inter-American Observatory in Chile on
18 August 2017 at 00:15:23 and detected a new source at right ascen-
sion α2000 = 13 h 09 m 48.07 s and declination δ2000 = −23° 22′ 53.7″,
not present in archival images of that galaxy (Fig. 2; see Methods for
a timeline of the merger and ensuing immediate follow-up). We are
one of a few groups who discovered the same source within 45 min of
each other (see Methods). It was first announced by the Swope team31,
who named it ‘SSS17a’, but here we use the official IAU designation,
AT 2017gfo.
Following the detection of this source, we initiated an intensive
follow-up campaign with LCO, obtaining multi-band images of
AT 2017gfo for several days, taken from each of our three Southern
Hemisphere sites (the Siding Spring Observatory in Australia, the
South African Astronomical Observatory, and the Cerro Tololo Inter-
American Observatory in Chile). AT 2017gfo was visible for less than
two hours each night owing to the proximity of its position on the sky to
the Sun, but having a multi-site observatory allowed us to obtain three
epochs of observations per 24-h period, capturing the rapid evolution
of the event (Fig. 3).
Our densely sampled light curve reveals that the optical transient
peaked approximately 1 day after the merger, followed by rapid fading
at a rate of about 2 mag per day in the g band, about 1 mag per day in
the r band, and about 0.8 mag per day in the i band. The rapid lumi-
nosity decline is unlike that of any supernova (Extended Data Fig. 4),
but is broadly consistent with theoretical predictions of kilonovae (see,
for example, refs 2 and 3). From the temporal and spatial coincidence
of this event with both a gravitational-wave signal from a binary neu-
tron-star merger and a short-duration γ-ray burst, we conclude that
AT 2017gfo is the kilonova associated with the same merger.
We first compare our observations to analytical models from the
literature. The short rise time and luminous bolometric peak of more
than 3 × 1041 erg s−1 (as indicated by blackbody fits to post-peak multi-
colour data; see Methods) are consistent with a low-opacity ejected
mass according to available analytical models11,32, but the observed
high early temperature is not (see Methods).
With this in mind, we compare the observations to detailed
numerical radiation transport models of kilonova light curves and
spectra21. The model parameters are the total ejecta mass, the charac­
teristic expansion velocity, defined as (2E/Mej)1/2 (where E is the total
kinetic energy imparted on the ejecta mass Mej), and the mass frac-
tion of lanthanide species, which are crucial in setting the opacity.
This model solves the multi-wavelength radiation transport equation
2 | NAT U R E | VO L 0 0 0 | 0 0 m o n t h 2 0 1 7
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
V + 1.5 i – 2.0 –20
g + 3.0 z – 3.5
14 r – 1.0 w
–18
16
Apparent magnitude
–16
Absolute magnitude
18
–14
20
–12
22
–10
24
0 1 2 3 4 5
Time since merger (rest-frame days)
Figure 3 | LCO light curves of the kilonova AT 2017gfo. Our rapid-
response high-cadence follow-up constrains the peak of the light curve
to approximately 1 day after the merger. Numerical radioactive-decay-
powered kilonova models21 are shown for an ejecta mass of 2 × 10−2M
(solid lines) and 2.5 × 10−2M (dashed lines), a characteristic ejecta
velocity of 0.3c and a low lanthanide fraction of 10−4.5. Error bars denote
1σ uncertainties. Data from the same site, filter and night are binned for
clarity. Magnitudes are corrected for host-galaxy contamination using
image subtraction, and for Milky Way extinction.
using detailed opacities derived from millions of atomic lines, while
self-consistently calculating the temperature and ionization/excitation
state of the radio­actively heated ejecta (see ref. 21 for more details).
This allows us to match the per-band light curves, rather than the
bolometric luminosity.
This approach produced a better match to our data, reproducing
most of the luminosity evolution (except in the g band; see below) using
an ejecta mass of (2–2.5) × 10−2M (where M is the solar mass), a
characteristic ejecta velocity of 0.3c (where c is the speed of light) and
a low lanthanide mass fraction of Xlan = 10−4.5 (Fig. 3), corresponding
to an effective opacity of κ  1 cm2 g−1 (similar parameters also fit our
optical spectra presented in ref. 33). This is evidence that the merger
produced a component of ejecta composed primarily of light (atomic
number A  140) r-process isotopes. In contrast, the lanthanide mass
fraction expected from the production of heavy r-process elements is
Xlan = 10−2–10−1 (ref. 34), corresponding to κ ≈ 10 cm2 g−1. A substan-
tial mass of ejecta must therefore have experienced substantial weak
interactions, owing to shock heating or neutrino interactions; these
interactions would have raised the proton-to-neutron ratio from its
initial value in the neutron star. In such a case, the neutrons available
for capture would be exhausted before nucleosynthesis could build up
a noticeable abundance of elements with A  140.
The discrepancy in the g band (and a smaller discrepancy in the
r band) may be due to a composition gradient in the ejecta (the model21
we used assumes a uniform composition). A radial gradient in the
lanthanide abundance, in which Xlan varies from about 10−6 in the
outermost layers to about 10−4 in the interior layers, could lead to faster
reddening of the emission21, which may fit the data better. Even more
lanthanide-rich ejecta (Xlan > 10−2) could be revealed through emission
at later times and redder wavelengths than covered by our data12–14.
Luminous infrared emission (J ≈ 17 mag, H ≈ 16 mag, Ks ≈ 15.5 mag;
although some of this emission may be contributed by the host galaxy)
is indeed found in observations taken 2.5 days and 3.5 days after the
merger35. It is possible that an additional source of radiation, perhaps
related to the γ-ray burst engine, contributes to the early blue emission,

and could provide an alternative explanation for the g- and r-band dis-
crepancies. Future modelling efforts will need to explore these options
and their effects on the predicted light curves.
The discovery of a kilonova coincident with gravitational waves
from a binary neutron-star merger and with a short burst of γ-rays
provides striking evidence in favour of the main theoretical picture of
neutron-star mergers. These detections confirm that binary
neutron-star mergers produce kilonovae with emission properties
broadly in agreement with theoretical predictions. Our early optical to
near-infrared light curve shows evidence for a lanthanide-poor compo-
nent of the mass ejected in the merger, and indications for a blue power
source in addition to radioactive decay. The rapid optical evolution
explains why transient surveys have so far not detected such events, but
the upcoming Large Synoptic Survey Telescope will detect the optical
emission of hundreds of kilonovae per year out to distances beyond
those accessible to current gravitational-wave detectors (see Methods).
Online Content Methods, along with any additional Extended Data display items and
Source Data, are available in the online version of the paper; references unique to
these sections appear only in the online paper.
received 12 September; accepted 21 September 2017.
Published online 16 October 2017.
1. Li, L.-X. & Paczyń ski, B. Transient events from neutron star mergers. Astrophys.
J. 507, L59 (1998).
2. Rosswog, S. Mergers of neutron star–black hole binaries with small mass
ratios: nucleosynthesis, γ-ray bursts, and electromagnetic transients. Astrophys.
J. 634, 1202–1213 (2005).
3. Metzger, B. D. et al. Electromagnetic counterparts of compact object mergers
powered by the radioactive decay of r-process nuclei. Mon. Not. R. Astron. Soc.
406, 2650–2662 (2010).
4. Wollaeger, R. T. et al. Impact of ejecta morphology and composition on the
electromagnetic signature of neutron star mergers. Preprint at https://arxiv.
org/abs/1705.07084 (2017).
5. Metzger, B. D. Kilonovae. Living Rev. Relativ. 20, 3 (2017).
6. Eichler, D. et al. Nucleosynthesis, neutrino bursts and γ-rays from coalescing
neutron stars. Nature 340, 126–128 (1989).
7. Narayan, R., Paczynski, B. & Piran, T. γ-ray bursts as the death throes of
massive binary stars. Astrophys. J. 395, L83 (1992).
8. Fong, W. & Berger, E. The locations of short γ-ray bursts as evidence for
compact object binary progenitors. Astrophys. J. 776, 18 (2013).
9. Hotokezaka, K. et al. Mass ejection from the merger of binary neutron stars.
Phys. Rev. D 87, 024001 (2013).
10. Kasen, D., Fernández, R. & Metzger, B. D. Kilonova light curves from the disc
wind outflows of compact object mergers. Mon. Not. R. Astron. Soc. 450,
1777–1786 (2015).
11. Grossman, D. et al. The long-term evolution of neutron star merger remnants.
II. Radioactively powered transients. Mon. Not. R. Astron. Soc. 439, 757–770
(2014).
12. Barnes, J. & Kasen, D. Effect of a high opacity on the light curves radioactively
powered transients from compact object mergers. Astrophys. J. 775, 18–26
(2013).
13. Kasen, D., Badnell, N. R. & Barnes, J. Opacities and spectra of the r-process
ejecta from neutron star mergers. Astrophys. J. 774, 25–37 (2013).
14. Tanaka, M. & Hotokezaka, K. Radiative transfer simulations of neutron star
merger ejecta. Astrophys. J. 775, 113–128 (2013).
15. Perley, D. A. et al. GRB 080503: implications of a naked short γ-ray burst
dominated by extended emission. Astrophys. J. 696, 1871–1885 (2009).
16. Tanvir, N. R. et al. A ‘kilonova’ associated with the short-duration γ-ray burst
GRB 130603B. Nature 500, 547–549 (2013).
17. Berger, E., Fong, W. & Chornock, R. An r-process kilonova associated with the
short-hard GRB 130603B. Astrophys. J. 774, L23 (2013).
18. Yang, B. et al. A possible macronova in the late afterglow of the long-short
burst GRB 060614. Nat. Commun. 6, 7323 (2015).
19. Jin, Z.-P. et al. The macronova in GRB 050709 and the GRB-macronova
connection. Nat. Commun. 7, 12898 (2016).
20. Abbott, B. P. et al. Multi-messenger observations of a binary neutron star
merger. Astrophys. J. 848, https://doi.org/10.3847/2041-8213/aa91c9
(2017).
21. Kasen, D. et al. Origin of the heavy elements in binary neutron-star mergers
from a gravitational-wave event. Nature http://doi.org/10.1038/nature24453
(2017).
22. LIGO Scientific Collaboration and Virgo Collaboration. GW170817: observation
of gravitational waves from a binary neutron star inspiral. Phys. Rev. Lett.
https://doi.org/10.1103/PhysRevLett.119.161101 (2017).
23. LIGO Scientific Collaboration et al. Advanced LIGO. Class. Quantum Gravity 32,
074001 (2015).
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Letter RESEARCH
24. Acernese, F. et al. Advanced Virgo: a second generation interferometric
gravitational wave detector. Class. Quantum Gravity 32, 024001 (2015).
25. Connaughton, V. Fermi GBM trigger 170817.529 and LIGO single IFO trigger.
GCN Circ. 21506 (2017).
26. LIGO Scientific Collaboration and Virgo Collaboration. LIGO/Virgo G298048:
further analysis of a binary neutron star candidate with updated sky
localization. GCN Circ. 21513 (2017).
27. Brown, T. M. et al. Las Cumbres Observatory Global Telescope Network. Publ.
Astron. Soc. Pacif. 125, 1031–1055 (2013).
28. Gehrels, N. et al. Galaxy strategy for LIGO-Virgo gravitational wave counterpart
searches. Astrophys. J. 820, 136–144 (2016).
29. Singer, L. P. & Price, L. R. Rapid Bayesian position reconstruction for
gravitational-wave transients. Phys. Rev. D 93, 024013 (2016).
30. Freedman, W. L. et al. Final results from the Hubble Space Telescope key
project to measure the Hubble Constant. Astrophys. J. 553, 47–72 (2001).
31. Coulter, D. A. et al. LIGO/Virgo G298048: potential optical counterpart
discovered by Swope telescope. GCN Circ. 21529 (2017).
32. Dietrich, T. & Ujevic, M. Gravitational waves and mass ejecta from binary
neutron star mergers: effect of the mass ratio. Class. Quantum Gravity 34,
105014 (2017).
33. McCully, C. et al. The rapid reddening and featureless optical spectra of the
optical counterpart of GW170817, AT 2017gfo, during the first four days.
Astrophys. J. (in the press).
34. Lippuner, J. & Roberts, L. F. r-process lanthanide production and heating rates
in kilonovae. Astrophys. J. 815, 82 (2015).
35. Kasliwal, M. et al. Illuminating gravitational waves: a concordant picture of
photons from a neutron star merger. Science http://doi.org/10.1126/science.
aap9455 (2017).
36. Arcavi, I. et al. Optical follow-up of gravitational-wave events with Las Cumbres
Observatory. Astrophys. J. (in the press).
Acknowledgements We are indebted to W. Rosing and the LCO staff for making
these observations possible, and to the LIGO and Virgo science collaborations.
We thank L. Singer, T. Piran and W. Fong for assistance with planning the LCO
observing program. We appreciate assistance and guidance from the LIGO–
Virgo Collaboration—Electromagnetic follow-up liaisons. We thank B. Tafreshi
and G. M. Árnason for helping to secure Internet connections in Iceland while
this paper was being reviewed. Support for I.A. and J.B. was provided by the
National Aeronautics and Space Administration (NASA) through the Einstein
Fellowship Program (via grant numbers PF6-170148 and PF7-180162,
respectively). G.H., D.A.H. and C.M. are supported by US National Science
Foundation (NSF) grant AST-1313484. D.P. and D.M. acknowledge support
by Israel Science Foundation grant number 541/17. D.K. is supported in part
by a Department of Energy (DOE) Early Career award DE-SC0008067, a DOE
Office of Nuclear Physics award DE-SC0017616, and a DOE SciDAC award
DE-SC0018297, and by the Director, Office of Energy Research, Office of High
Energy and Nuclear Physics, Divisions of Nuclear Physics, of the US Department
of Energy under contract number DE-AC02-05CH11231. This research used
resources of the National Energy Research Scientific Computing Center, a
DOE Office of Science User Facility supported by the Office of Science of the
US Department of Energy under contract number DE AC02-05CH11231. This
research has made use of the NASA/IPAC Extragalactic Database, which is
operated by the Jet Propulsion Laboratory, California Institute of Technology,
under contract with NASA. The Digitized Sky Surveys were produced at
the Space Telescope Science Institute (STScI) under US Government grant
number NAG W-2166. The UK Schmidt Telescope was operated by the Royal
Observatory Edinburgh, with funding from the UK Science and Engineering
Research Council (later the UK Particle Physics and Astronomy Research
Council), until June 1988, and thereafter by the Anglo-Australian Observatory.
Supplementary funding for sky-survey work at the STScI is provided by the
European Southern Observatory.
Author Contributions I.A. is Principal Investigator of the LCO gravitational-wave
follow-up program; he initiated and analysed the observations presented here
and wrote the manuscript. G.H. helped with the LCO alert listener and ingestion
pipeline, with follow-up observations and image analysis, and performed the
blackbody fits. D.A.H. is the LCO–LIGO liaison, head of the LCO supernova
group, and helped with the manuscript. C.M. assisted with obtaining and
analysing data, and helped with the LCO alert listener. D.P. helped design the
LCO follow-up program, assisted with the galaxy prioritization pipeline and
contributed to the manuscript. D.K. and J.B. developed theoretical models and
interpretations. M.Z. built the galaxy prioritization pipeline. S. Vasylyev built the
LCO alert listener and ingestion pipeline. D.M. helped in discussions and with
the manuscript. S. Valenti helped with image analysis and with the manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial
interests. Readers are welcome to comment on the online version of the paper.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations. Correspondence and
requests for materials should be addressed to I.A. (arcavi@gmail.com).
Reviewer Information Nature thanks R. Chevalier, C. Miller and the other
anonymous reviewer(s) for their contribution to the peer review of this work.
0 0 M o n t h 2 0 1 7 | VO L 0 0 0 | NAT U R E | 3
 RESEARCH Letter
METHODS
Gravitational wave follow-up strategy and kilonova discovery. The Las Cumbres
Observatory (LCO)27 consists of 20 telescopes (two 2-m, nine 1-m and nine 0.4-m
in diameter) at six sites around the world, operated robotically as one network
using dynamical scheduling software. As stated in the main text, we use a galaxy-
targeted follow-up strategy rather than a tiling one28. Our galaxy selection strategy
prioritizes galaxies that are at higher-probability locations and distances in the
gravitational-wave localization region26, that have a higher intrinsic B-band lumi-
nosity (indicative of higher mass), and in which LCO is more likely to be sensitive
to a kilonova. More details are provided in ref. 36. The timeline of the discovery,
immediate follow-up and the visibility of NGC 4993 are depicted in Extended Data
Fig. 1. In addition to our detection, AT 2017gfo was independently detected by the
Swope, DECam, DLT40, MASTER and VISTA groups31,37–43,64.
Photometry. Images from the LCO 1-m telescopes were pre-processed using the
Python-based BANZAI pipeline. Photometry was then extracted using the PyRAF-
based LCOGTsnpipe pipeline44 by performing image subtraction45 followed by
point spread function fitting. We use images taken after the kilonova faded below
our detection limits as subtraction references. Our V-band data are calibrated to
the AAVSO Photometric All-Sky Survey46 in the Vega system, grizw-band data are
calibrated to the AB system using Sloan Digital Sky Survey (SDSS) fields observed
on the same night as AT 2017gfo, with the w band (which is a broad g + r + i
band) treated as an r band. We correct all photometry for Milky Way extinction47
retrieved via the NASA/IPAC Extragalactic Database (http://ned.ipac.caltech.edu/).
We adopt a Tully–Fisher distance of 39.5 Mpc (distance modulus of 32.98 mag)30
to NGC 4993 retrieved via the NASA/IPAC Extragalactic Database.
Blackbody fits. Kilonovae are expected to display roughly blackbody emission
(perhaps with a steeper fall-off at short wavelengths due to line blanketing10,12,13,33).
We fitted a blackbody spectrum to each epoch containing data in more than two
bands (excluding w-band data) using Markov chain Monte Carlo (MCMC) simu-
lations through the Python emcee package48 (Extended Data Fig. 2). We find that
the photospheric radius remains roughly constant during the first few days after
peak at a value of about 5 × 1014 cm while the temperature declines from about
6,500 K 1.4 days after the peak to about 4,000 K 2.5 days after the peak (Extended
Data Fig. 3). We calculate the bolometric luminosity of the blackbody and take that
to be the bolometric luminosity of the event.
Comparison to supernova light curves. AT 2017gfo peaks at an absolute magni-
tude that is fainter than most supernovae, but comparable to that of some type IIb
supernovae, and to plateau luminosities of type IIP supernovae (see, for example,
ref. 49). However, AT 2017gfo evolves faster than any known supernova.
In Extended Data Fig. 4 we compare it to standard type Ia and type Ib/c light
curves50,51, as well as to some of the most rapidly evolving supernovae known52,53,
SN 2002bj and SN 2010X. We also plot the plateau drop phase of the prototypical
type IIP supernova54 SN 1999em. Type IIP supernova light curves have an approxi-
mately 100-day plateau, followed by a rapid drop in luminosity as the power source
changes from shock heating to radioactive decay of 56Co. Still, this sharp decline
is slower than the decline in AT 2017gfo. In Extended Data Fig. 4 we also plot the
DLT40 and ATLAS non-detection pre-discovery limits55,56 of AT 2017gfo, which
further rule out a type IIP supernova origin.
Fits to analytical kilonova models. The basic predictions for the peak time, lumi-
nosity and temperature of a kilonova, assuming a spherically symmetric, uniform
mass distribution for an ejecta in homologous expansion, are11:
 Mej, −2 κ10 1 / 2
tpeak ≈ 4.9 days ×   (1)
 vej, −1 
 Mej, −2 κ10 −α / 2
L peak ≈ 2.5 × 10 40 erg s−1 × Mej, −2   (2)
 vej, −1 
α / 8 −(α + 2) / 8 (α − 2) / 8
Tpeak ≈ 2, 200 K × M −
ej, −2 κ10 v ej, −1 (3)
where Mej,−2 is the ejecta mass in units of 10−2M, κ10 is the opacity of the ejecta
mass in units of 10 cm2 g−1, vej,−1 is the ejecta velocity in units of 0.1c, and α is the
power-law index that describes the time dependence of the energy emitted by
­radioactive decay. Here we use α = 1.3, which is typically assumed for ­r-process
decay57. The peak luminosity (see equation (2)) is approximately 1,000 times
brighter than a nova, giving kilonovae their name3 (although some use the more
general name ‘macronovae’)58.
These simple relations can reproduce the short rise time and bright peak lumi-
nosity deduced from the blackbody fits (Extended Data Fig. 5) with an ejecta mass
Mej of a few hundredths of solar masses and a low (κ  1 cm2 g−1) opacity. However,
using these values does not reproduce the observed colours, as it underpredicts
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
the observed temperature (see equation (3)). We use these parameters as starting
points for MCMC simulations to fit more sophisticated analytical models32 based
on approximations to numerical relativity simulations. We fitted the models to the
bolometric light curve rather than using the model bolometric corrections to fit
the per-band light curves, since the corrections are only valid for times
>2 days × (10−2M/Mej)−1/3.2 after the merger, which would miss much of our
data. We fix the heating rate coefficient ε 0 = 1.58 × 1010 erg g−1 s−1 and leave the
ejecta mass (Mej), the minimum and maximum ejecta velocities (vej,min and vej,max),
the opacity (κ), and the geometrical parameters (θej and Φej) as free parameters.
We use the public code provided in ref. 59 for these models and adopt the time-
varying thermalization efficiency found in ref. 60. Our MCMC fits converge on an
ejecta mass of (4.02 ± 0.05) × 10−2M (1σ uncertainties), but do not constrain the
ejecta velocities (Extended Data Fig. 6) or the geometrical parameters (in
ref. 59 it is demonstrated that in general the geometrical parameters cannot be
constrained in this model). We compare the individual band magnitudes from this
fit, using the bolometric corrections supplied by the model and find that they are
redder than the observations. We conclude that even the more sophisticated ana-
lytical models32 (under the stated assumptions for α, ε 0 and the thermalization
efficiency) cannot reproduce the colour evolution of our event. As stated for our
numerical models21 in the main text, a composition (and hence opacity) gradient
or an additional power source, could explain the colour-evolution discrepancy.
Rates. Given the light curve properties reported in the main text, we can explore
how many AT 2017gfo-like events are expected to be seen by different optical
transient surveys, without relying on a gravitational-wave trigger. The number of
kilonovae per year N potentially seen in E epochs by a survey covering a fraction f
of the sky down to limiting magnitude L and with cadence C days is:
N = fR × 100.6(L −mC (E− 1) − m p) (4)
where R is the rate of kilonovae per year on the entire sky out to a distance d, Δm
is the decline rate of the kilonova in magnitudes per day, and mp is the apparent
peak magnitude of the kilonova at distance d (we ignore time dilation effects from
an expanding Universe). Using the values from our r-band data (mp = 17, Δm = 1)
and assuming R = 1, we plot the number of detectable kilonovae in Extended Data
Fig. 7. We find, for example, that a survey with a limiting magnitude of 21 and
sky coverage of 4,000 square degrees with 3-day cadence (similar to the Palomar
Transient Factory61,62) would have a two-epoch detection of only one kilonova
roughly every 2–3 years. The upcoming Large Synoptic Survey Telescope,
reaching a magnitude of 24 on roughly half of the sky with 3-day cadence,
could obtain three epochs for one kilonova per year, and two epochs for each of
100 kilonovae per year. Equation (4) demonstrates that increasing the cadence of a
survey has a larger effect on kilonova detections than increasing the sky coverage.
It is therefore likely that the Large Synoptic Survey Telescope could discover even
more kilonovae in its ‘deep drilling’ fields.
Data availability. The photometric data that support the findings of this study are
available in the Open Kilonova Catalog63, https://kilonova.space. Source Data for
Fig. 3 are provided with the online version of the paper.
37. Allam S. et al. LIGO/Virgo G298048: DECam optical candidate. GCN Circ.
21530 (2017).
38. Yang, S. et al. LIGO/Virgo G298048: DLT40 optical candidate. GCN Circ. 21531
(2017).
39. Tanvir, N. R. & Levan, A. J. LIGO/Virgo G298048: VISTA/VIRCAM detection of
candidate counterpart. GCN Circ. 21544 (2017).
40. Lipunov, M. et al. LIGO/Virgo G297595: MASTER observations of the NGC
4993. GCN Circ. 21546 (2017).
41. Coulter, D. A. et al. Swope Supernova Survey 2017a (SSS17a), the optical
counterpart to a gravitational wave source. Science http://doi.org/10.1126/
science.aap9811 (2017).
42. Soares-Santos, M. et al. The electromagnetic counterpart of the binary neutron
star merger LIGO/VIRGO GW170817. I. Discovery of the optical counterpart
using the dark energy camera. Astrophys. J. 848, https://doi.
org/10.3847/2041-8213/aa9059 (2017).
43. Valenti, S. et al. The discovery of the electromagnetic counterpart of
GW170817: kilonova AT 2017gfo/DLT17ck. Astrophys. J. 848, https://doi.
org/10.3847/2041-8213/aa8edf (2017).
44. Valenti, S. et al. The diversity of type II supernovae versus the similarity in their
progenitors. Mon. Not. R. Astron. Soc. 459, 3939–3962 (2016).
45. Becker, A. HOTPANTS: High Order Transform of PSF ANd Template
Subtraction. Astrophys. Source Code Lib. ascl:1504.004 (2015).
46. Henden, A. A. et al. The AAVSO Photometric All-Sky Survey (APASS). AAS Meet.
Abstr. 214, 407.02 (2009).
47. Schlafly, E. F. & Finkbeiner, D. P. Measuring reddening with Sloan Digital Sky
Survey stellar spectra and recalibrating SFD. Astrophys. J. 737, 103–115 (2011).
48. Foreman-Mackey, D., Hogg, D. W., Lang, D. & Goodman, J. emcee: the MCMC
Hammer. Publ. Astron. Soc. Pacif. 125, 306–312 (2013).
49. Arcavi, I. Hydrogen-rich core collapse supernovae. In Handbook of Supernovae
(eds Alsabti, A. W. & Murdin, P.) (Springer, in the press).

50. Conley, A. et al. SiFTO: an empirical method for fitting SN Ia light curves.
Astrophys. J. 681, 482–498 (2008).
51. Taddia, F. et al. Early-time light curves of type Ib/c supernovae from the SDSS-II
supernova survey. Astron. Astrophys. 574, A60 (2015).
52. Poznanski, D. et al. An unusually fast-evolving supernova. Science 327, 58–60
(2010).
53. Kasliwal, M. M. et al. Rapidly decaying supernova 2010X: a candidate “.Ia”
explosion. Astrophys. J. 723, L98 (2010).
54. Leonard, D. C. et al. The distance to SN 1999em in NGC 1637 from the
expanding photosphere method. Publ. Astron. Soc. Pacif. 114, 35–64 (2002).
55. Yang, S. et al. LIGO/Virgo G298048: continued observation for DLT17ck. GCN
Circ. 21579 (2017).
56. Tonry J. et al. LIGO/Virgo G298048: ATLAS pre-discovery limits 601 to 16 days
before first detection of SSS17a/DLT17ck. GCN Circ. 21886 (2017).
57. Korobkin, O., Rosswog, S., Arcones, A. & Winteler, C. On the astrophysical
robustness of the neutron star merger r-process. Mon. Not. R. Astron. Soc. 426,
1940–1949 (2012).
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Letter RESEARCH
58. Kulkarni, S. R. Modeling supernova-like explosions associated with
gamma-ray bursts with short durations. Preprint at https://arxiv.org/abs/
astro-ph/0510256 (2005).
59. Coughlin, M. et al. Towards rapid transient identification and characterization
of kilonovae. Preprint at https://arxiv.org/abs/1708.07714 (2017).
60. Barnes, J., Kasen, D., Wu, M.-R. & Martínez-Pinedo, G. Radioactivity and
thermalization in the ejecta of compact object mergers and their impact on
kilonova light curves. Astrophys. J. 829, 110–129 (2016).
61. Law, N. M. et al. The Palomar Transient Factory: system overview, performance
and first results. Publ. Astron. Soc. Pacif. 121, 1395–1408 (2009).
62. Rau, A. et al. Exploring the optical transient sky with the Palomar Transient
Factory. Publ. Astron. Soc. Pacif. 121, 1334–1351 (2009).
63. Guillochon, J. et al. An open catalog for supernova data. Astrophys. J. 835, 64
(2017).
64. Lipunov, V. et al. MASTER optical detection of the first LIGO/Virgo NSs merging
GW170817/G298048. Astrophys. J. (in the press).
 RESEARCH Letter

4 ' %! 

 0

0 


  & '%$($

 !"#$ % %% '%!)  0

"$$') 1) * 2 - # %3

 *+ & '"$ $ "
 
 

 

 

 
 0




 

 
,-$ % &-"
,-$ % . '$- /! ,-$ % '$"




  
   










Extended Data Figure 1 | Timeline of the discovery and the
observability of AT 2017gfo in the first 24 h following the merger.
The curved lines denote the airmass and altitude (in degrees above the
horizon) of the position of AT 2017gfo on the sky at each LCO Southern
Hemisphere site from the start of the night until the hour-angle limit
of the LCO 1-m telescopes. The vertical thick lines denote the times
when LCO images were obtained (colours correspond to the different
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
filters as denoted in the legend of Fig. 3). AT 2017gfo was observable for
approximately 1.5 h at the beginning of the night. Having three Southern
Hemisphere sites allowed us to detect the kilonova approximately 6.5 h
after the LIGO-Virgo localization, follow it approximately 10 h later, and
continue to observe it three times per 24-h period for the following days
(Fig. 3). Counterpart announcement is from ref. 31.
 Letter RESEARCH

Extended Data Figure 2 | Blackbody fits. MCMC parameter distributions
(a–f) and spectral energy distributions (luminosity density Lλ as a
function of wavelength) with the blackbody fits (g–l) are shown for the
six epochs (noted by their modified Julian dates, MJD) with observations
in more than two bands after excluding w-band data. In the parameter
distributions, contour lines denote 50% and 90% bounds, the red and blue
solid lines overplotted on each histogram denote the mean and median of
each parameter distribution (respectively), and the dashed lines denote
68% confidence bounds. Error bars on the luminosity densities denote 1σ
uncertainties.

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH Letter

Extended Data Figure 3 | Bolometric luminosity, photospheric radius
and temperature deduced from blackbody fits. Error bars denote 1σ
uncertainties (n = 200). The large uncertainties in the later epochs might
be due to a blackbody that peaks redward of our available data, so these
data points should be considered to be temperature upper limits. Our
MCMC fits of an analytical model32 to the bolometric luminosity are
shown in blue, and the numerical models21 from Fig. 3 are shown in red in
the top panel. The numerical models were tailored to fit Vriw bands, but
not the g band, which is driving the high bolometric luminosity at early
times.

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 Letter RESEARCH

Extended Data Figure 4 | AT 2017gfo evolves faster than any known
supernova, contributing to its classification as a kilonova. We compare
our w-band data of AT 2017gfo (red; arrows denote 5σ non-detection
upper limits reported by others55,56) to r-band templates of common
supernova types (types Ia and Ib/c normalized to peaks of −19 mag
and −18 mag, respectively)50,51, to r-band data of two rapidly evolving
supernovae52,53 (SN 2002bj and SN 2010X) and to R-band data of the drop
from the plateau of the prototypical type IIP supernova54 SN 1999em
(dashed line; shifted by 1 mag for clarity).

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH Letter

Extended Data Figure 5 | Peak luminosity and time of AT 2017gfo velocities vej (red and blue lines). The rise time and peak luminosity of
compared to simple analytical predictions. The parameters11 from AT 2017gfo (black arrow) can be reproduced by an ejecta velocity vej ≈ 0.3c
equations (1) and (2) are shown for different values of the ejecta mass and a low opacity of κ  1 cm2 g−1. Matching the data with higher opacities
Mej (solid lines), the opacity κ (dashed lines), and for two different ejecta would require higher ejecta velocities.

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

Extended Data Figure 6 | Parameter distribution for MCMC fits of
analytical kilonova models32 to our bolometric light curve. The contour
lines denote 50% and 90% bounds. The red and blue solid lines overplotted
on each histogram denote the mean and median of each parameter
distribution (respectively). The dashed lines denote 68% confidence
bounds. The fits converge on an ejecta mass of (4.02 ± 0.05) × 10−2M
but they do not constrain the velocity (converging on the largest possible
range) or the geometrical parameters (θej and Φej), nor do they reproduce
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
Letter RESEARCH
the colour evolution of our event (not shown). This indicates that these
models may not be entirely valid for AT 2017gfo (although in ref. 59 it
is shown that the geometrical parameters cannot be constrained either
way). Our numerical models21, on the other hand, which include detailed
radiation transport calculations, do provide a good fit to the data (Fig. 3)
with Mej = (2–2.5) × 10−2M, vej = 0.3c, and a lanthanide mass fraction of
Xlan = 10−4.5, corresponding to an effective opacity of κ  1 cm2 g−1.
 RESEARCH Letter

Extended Data Figure 7 | Expected kilonova rates in optical transient to the entire sky, and should be multiplied by the fraction of sky covered by
surveys. The number of AT 2017gfo-like events per year detectable by the survey. We assume that the intrinsic rate of events is one per year out
r-band transient surveys in two (solid lines), three (dashed lines) and five to 40 Mpc (scaling accordingly to larger distances).
(dotted lines) epochs before fading from view. The numbers of events refer

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.